Skip to content

v0.1.6 — why we call it a Bible

Latest

Choose a tag to compare

@tonydzi tonydzi released this 05 Sep 04:12

Docs only, but it closes a gap the repo had carried since July: the argument for the name was never actually written down.

docs/why-a-bible.md makes it. The design problem underneath the whole repo is that many agents run on many machines, they do not see each other, and no single participant — human or model — holds the system in its head at any moment. What makes them one system rather than six robots is a shared text that every actor reads before acting. That is Harari's coordination-of-strangers framing, taken only in its uncontroversial half and paraphrased rather than quoted: we claim the function, none of the theology.

Hence the word. Not "config", which implies a machine consumes it; not "prompt", which implies a session forgets it. A versioned codex that humans and agents both load, with explicit precedence rules and a known ritual for overturning a rule. Constitutional AI is the obvious prior art for governing a model with a written document, and the page says so plainly — this is the analogous move one level up, governing a fleet and its humans rather than one model's outputs.

The page also states the limits out loud, because a lone founder plus a religious metaphor is a well-known failure genre. This is a reference implementation we run ourselves on six machines. It is not "proven safe" and it is not deployed at N companies. External reproductions are still being counted from zero — which is the point of publishing, since the harness exists so strangers can check the claims without trusting us. A failed reproduction is worth more to us than a star.

Also in this release: GitHub URLs repointed from the deleted org to tonydzi, so the cross-links in README, ROADMAP and the devlog resolve again.